I built oosmetrics to answer a question I kept running into: which repos are actually gaining momentum right now, not just which ones already have the most stars?
It tracks 300,000+ GitHub repos continuously and computes three metrics per repo:
- Growth rate (7d and 30d): star velocity, not raw count
- Acceleration: is growth speeding up, stable, or declining?
- Originality: how does this repo perform relative to its taxonomy peers? A 2K-star Rust HTTP framework that leads its niche scores higher than a 10K-star repo that's average for its category
Each metric gets a percentile-based letter grade (S to F). The feed is filterable by topic, language, and sort order, with a "Rising Stars" view for repos that entered tracking recently but are already accelerating.
A few technical details that might interest HN:
- GitHub's Search API caps results at 1,000 per query. Discovery works around this by dynamically splitting star-count brackets when a bracket hits the ceiling; currently at 315 brackets
- Processing is fully continuous. A worker pool picks any repo not processed in the last 24h and runs snapshot -> taxonomy -> metrics -> grades -> cache invalidation
- Grades are percentile-based within the corpus (or within taxonomy for originality), so they recalibrate automatically as the dataset grows
- Repo pages include an on-demand AI overview (what it does, who it's for, quick start, license) generated from the README by a self-hosted language model and cached permanently, only generated on first visit
Everything is free, no login required. Would love feedback on the metric definitions and whether the grading feels calibrated.
Comments URL: https://news.ycombinator.com/item?id=47423780
Points: 2
# Comments: 3